End-to-End Dementia Status Prediction from Brain MRI Using Multi-task Weakly-Supervised Attention Network.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Computer-aided prediction of dementia from brain MRI aids in assessing disease stage and progression.
• Current methods often use preselected brain regions, potentially missing individual variations in dementia.
• These regions are typically standardized, neglecting subject-specific pathological changes.

Purpose of the Study:

• To develop a novel AI model for predicting dementia clinical scores from brain MRI.
• To incorporate individual subject specificities in dementia progression into the prediction model.
• To enable end-to-end feature learning and multi-task regression for dementia assessment.

Main Methods:

• Proposed a multi-task weakly-supervised attention network (MWAN).
• MWAN utilizes a dementia attention block to automatically identify subject-specific discriminative MRI locations.
• Employed cross-validation on ADNI-1 and ADNI-2 datasets for evaluation.

Main Results:

• The MWAN method demonstrated strong performance in predicting clinical dementia scores.
• The model successfully performed weakly-supervised discriminative localization in brain MR images.
• Experimental results validated the effectiveness of considering individual specificities.

Conclusions:

• The proposed MWAN method accurately predicts dementia clinical scores from brain MRI.
• The approach effectively identifies subject-specific brain regions relevant to dementia.
• This method offers a promising tool for dementia assessment and progression prediction.